Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

[ABCTG]

Twitter list of FSD beta
https://twitter.com/i/lists/1319121975265067009

Borat damages aTesla Model X
Watch Borat Crash Mail Truck Into Tesla Model X Door For Laughs

The next Mike Tyson fight is scheduled for Nov 28th
Tyson On Triller

 

[amolina]

That's right. 30+ years ago I was very annoyed about it predicting I could 10 or 20 fold the money if I could work with it. No way to do that of course which is ridiculous as in average looking back bonds don't bring you what stocks do and I assumed they invested in bonds. I learned later that there is no money to invest in the German pension funds at all, it's spend already and your pension slip is just a promise on paper.

Some say that explains why the rather prefer to keep that commitment low ....

Divesting in Tesla makes for the next 5-10 years no sense unless something really unexpected happens I have not yet thought about. It depends of course on your personal situation and should not be generalized. If the company is managed like now and continues to innovate and follow the mission you may extend that period to the far future. The question you need to ask is if there are other investments that have a better value - risk ratio than Tesla. SpaceX could be one IMO.

For now, there is 0 reason to waste any time on a devastation strategy

Yep. Even though I could increase my payment to my german pension by paying X Euros,( My employer would pay 2X Euros because it saves them taxes) , the return is still so shitty when I retire in the next 30 years. So I only pay the minimum that is required by law and invest it rather into my own long term index called TSLA.

Also, the tax consultants I spoke to, couldn't fathom that German state has a lot of risks ahead of itself with a massive undergoing transformation in the automobile industry. He almost laughed at me when I asked what if the German govt. is not able to give me the returns they promise.

On another note, I am not planning to divest from TSLA at all. I view it as my S&P index for the long term. I just think maybe I would have to sell and realize the capital gains I have enjoyed this year to convert it into LEAPS

 

[dw4ngg]

England to undergo lockdown for a month. Futures should be pretty bleak.

 

[kbM3]

Wow! I wonder if the capture & "solve" rate would slow, as cases get edgier and edgier...i.e. it's easier to pick the low hanging fruit, so to speak. So could 6 months turn into a year or two?.

Yes. This really becomes the central question. Exactly how do edge cases explode as you strive for lower intervention probability, and I suppose how much can you solve weird edge cases with a generalized solution. E.g. Hopefully after training on enough road debris, it can handle a school desk even though one has never been encountered. Although it could easily be millions/billions of miles between encountering school desks in the road, there are so many possibilities of strange road debris, that there is a good probability of encountering some weird situation akin to that, every 100k miles.

 

[Tim S]

Human drivers on a robo-only highway will be identified by crashing. If they survive, they will be jailed or heavily fined.

Maybe a transition step to robo-only highways will be robo-only lanes on human highways -- like a carpooling lane, but with closely-spaced robocars whipping past the human traffic at twice the speed. That will be good advertising for robocars and will teach even cocky human drivers that they really can't survive on a robo-only highway.

Yes. I met a guy about a year ago who was the founder and ceo of a local software company that was working with Tesla. He was excited about the beta software he had in his Model S. He told me that in 4 or 5 years, the left lane on US highways would be reserved for "smart vehicles" only. They would communicate with each other and convoy at speeds up to 100 mph. He seemed very confident this would happen.

 

[aronth5]

Yes. I met a guy about a year ago who was the founder and ceo of a local software company that was working with Tesla. He was excited about the beta software he had in his Model S. He told me that in 4 or 5 years, the left lane on US highways would be reserved for "smart vehicles" only. They would communicate with each other and convoy at speeds up to 100 mph. He seemed very confident this would happen.

Can or will it happen? Maybe, but does anyone think all the government agencies could get together to approve this in 4-5 years. Zero probability of that happening never mind getting the car companies to play together.

 

[EVortex]

Can or will it happen? Maybe, but does anyone think all the government agencies could get together to approve this in 4-5 years. Zero probability of that happening never mind getting the car companies to play together.

There may only be one car company left by then and it does not start with G, F, D, M, B, or V - and especially not N.

 

[AZRI11]

Good news from Shanghai https://twitter.com/Louisbotelho17/status/1322628441708048385?s=19 7 days a week two shifts a day.

 

[Lessmog]

There may only be one car company left by then and it does not start with G, F, D, M, B, or V - and especially not N.

Aha! Could it be T then? Tata! Tatra? Simca? Citroën/Rénault? Yugo? Zil?
Not Messerschmitt, not MG, not Baidu, not Volvo -- or is it Polestar? No, thats Geely. Lloyd? Scania?
I give up.

Edit: No, wait: Inzile! Of course.

 

[H Mak]

This implies (to me at least) that the vehicles will need some way to identify a robo vs human driver in other vehicles. If it’s robo, you can convoy closely. Otherwise give them a wider berth

Stick one of these on the back

 

[Lessmog]

Stick one of these on the back

Get in touch with : https://twitter.com/enn_nafnlaus/status/1321450235638546432

 

[PeterJA]

Yes. I met a guy about a year ago who was the founder and ceo of a local software company that was working with Tesla. He was excited about the beta software he had in his Model S. He told me that in 4 or 5 years, the left lane on US highways would be reserved for "smart vehicles" only. They would communicate with each other and convoy at speeds up to 100 mph. He seemed very confident this would happen.

I agree, but the guy might be disappointed to find the smart vehicles will communicate with each other using the same sensors they use for other driving -- cameras, radar and ultrasound. When each vehicle checks the speed and direction of the one ahead of it every microsecond, the entire convoy will react essentially as a unit to any obstacles on the road. Extra hardware and software for communication between vehicles will be unnecessary... just as lidar is unnecessary.

 

[JohnnyEnglish]

It is possible that the FSD beta is being trained on a partially built Dojo.
Final Dojo can apparently train 3 times per day.
Yes, in theory that could mean daily updates.

The other thing that might be different is they can just update the NN weights produced by training without a software update.


Prior to Dojo training takes longer. 7 days?

If this is true, as Dojo is progressively built, training will accelerate, improved versions will come faster.

I am pretty sure that Elon referred to the fact that they had built a prototype version of Dojo (using FPGAs?) ahead of them having the custom silicon they will use for what the video called 'final' Dojo. I am guessing the custom silicon is actually Dojo V1 and will improve iteratively over time, just like the inference hardware in Tesla cars - HW2, HW2.5, HW3 and HW4 in the next iteration.

The aim of Dojo is undoubtedly to improve training speed for NNs. However the statement in the video that Dojo would be able to train 3 times per day seems to misunderstand the underlying process. My understanding is that you would only want to train a NN once with your current set of training data. Once that has been done you would only train again when you have a different (enhanced) set of training data or the NN has been modified. I am guessing that the aim is to get training times down to less than 12 hours so that training can be completed overnight ready for the team to start review and test of the outputs the following day.

 

[Prunesquallor]

Weekend OT
RIP

 

[hobbes]

Good news from Shanghai https://twitter.com/Louisbotelho17/status/1322628441708048385?s=19 7 days a week two shifts a day.

Even better, but not exactly news :
They started doing 3 shifts after Chinese new year holidays:
Tesla Giga Shanghai roars back into action in pursuit of record Q4 deliveries
Tesla Giga Shanghai Will Increase to 3 Shifts After Chinese National Holiday

 

[Green Pete]

Even better, but not exactly news :
They started doing 3 shifts after Chinese new year holidays:
Tesla Giga Shanghai roars back into action in pursuit of record Q4 deliveries
Tesla Giga Shanghai Will Increase to 3 Shifts After Chinese National Holiday

Chinese new year is 3 months away. So "will start" not "started"

Edit: my bad. The confusion was that the holiday in question was not Chinese New year, so they have actually already started 3 shifts

National Day of the People's Republic of China - Wikipedia

Is not

Chinese New Year - Wikipedia


It's probably worth everyone knowing the difference as the shanghai factory will likely be shut for 3ish weeks around actual chinese New year, which is a totally normal occurance for all Chinese companies. But I'm sure FUD next spring about it being because of something more serious will be rampant

 

[Maitri982]

So based on the Dojo video a couple of posts back, they run the training computer 2 or 3 times per release. With Dojo at full speed daily updates

As a possiblist (but having learned only from the many videos), it seems hard not to imagine a pretty compelling FSD in 12-24months, at least in reasonably well regulated countries.

I wonder if you could count (will we see?) a "days per nine" count?

You should not interpret elons statement as implying they are training the computer with each update. There is still other code Coexisting with the hydranet that is easier to tweak and I imagine these updates are frequently just that.

 

[Maitri982]

1. Using Dojo as opposed to the existing trainer they may be moving from running training in 72h to running [the same] training in 7h, but
2. They are not running the same training, if only because they are shifting from '2.5D' to '4D'. and we are unsure if the 72h > 7h is a like-for-like comparison or also takes the 2.5>4D shift into account, and
3. But anyway as they shift more and more of the driving task into the central network, as opposed to being in separate modules, then the training task becomes larger. At least it does until a portion becomes sufficiently solved that it can be set aside for a while (i.e. training sessions will concentrate on other aspects until they too are sufficiently solved, then they will iterate around the mission space picking up edge cases in some sort of priority sequence).
4. Dojo shouldn't be thought of as being one monolithic computer. For sure it may (??) be capable of operating in that manner, but more likely it will be better to think of it as several computers each of which addresses part of the training process, e.g. self-labelling; running training; edge-case identification; etc.
5. It is unlikely that they will ever just dump new netweights in during a software update. They would almost always want to restructure the net.
6. But bear in mind it is over 30-years since I last worked on NN controllers (boy does that make me feel old), but I don't think the issues have changed (because we could see these issues even back then).
7. It is great to see all this coming to fruition.
8. Have a think about this for a moment. For about the next 10-years, the essential barrier to entry in the FSD game - quite apart from amassing the required data sets and feedback loops - is the ability to design/build/operate one of the world's largest supercomputers, either using mission-specific chipsets or paying extra for the overhead of generic chipsets. At any given moment there are probably only about a couple of dozen organisations able to play in that league, and only a handful of them will be in the first division, and most of them have other objectives (aerospace, nuclear, whatever) and there are only so many people that are viable hires for these organisations. The barrier to entry for competitors will likely get bigger, not smaller, for the next ten years (as competitors need to be comparable to Tesla before being able to credibly launch in the FSD marketplace).

regards, dspp/pp

To point 5...

I doubt they are doing that much knob turning of hyperparameters at this point. So weight updates would be the main thing if they push a network update out.

 

[wipster]

<snip>

I believe that this is the case at least in Florida but it doesn't matter. Once the software is ready, legislation will follow suit. Tesla flipping the switch won't come totally out of the blue, though. The capabilities will gradually improve and many owners will see stretches of thousands of miles without taking over once. A couple of iterations before robust statistics prove that it is safer than humans, many will be yelling to release FSD already.

Somewhere along the way between early Beta testers constantly on the edge and a car that can be trusted to do its job will be a twilight zone of not-quite-there-yet and complacent drivers that consider themselves passengers. I have no idea how this will pan out.

I have only been there once, making a presentation to PGA HQ in another life, but I swear the highways in West Palm Beach, FL might just present the biggest challenge for 4D FSD between 90 year old men (and women) who drive 40 in a 70 zone and young rich Lamborghini and Ferrari owners who drive over 100. FSD is going to have a LOT of learning to do down there, and if it can drive there, I believe it can learn anywhere.

Way worse than Chicago at rush hour... at least there everybody drives the minimum speed of 90 mph.

 

[Singuy]

England to undergo lockdown for a month. Futures should be pretty bleak.

Seems like these lock downs are lock down lite. Restaurants and bars only it seems.